ADAPTIVE SENSOR FUSION METHOD FOR AGRICULTURAL AND
ENVIRONMENTAL MONITORING
M. A. Dota
Institute of Exact and Natural Sciences Institute
University of Mato Grosso – UFMT
Rondonópolis – MT, Brazil
C. E. Cugnasca
Computer and Digital Systems Engineering Department
School of Engineering, University of São Paulo – USP
São Paulo – SP, Brazil
ABSTRACT
Environmental and agricultural monitoring involves continuous observation in
areas such as grains crop, in order to evaluate changes in the environment.
Wireless Sensor Networks may be employed in this context allowing the use of
several sensors to collect data. The aim is to present the design of an Adaptive
Sensor Fusion method appropriate for agricultural and environmental monitoring
applications through Wireless Sensor Networks. The proposed method integrates
Adaptive Technology mechanisms to enable it to dynamically adapt to changes
that may occur in real time resulting in better data analysis and processing by
fusion techniques and, consequently, in a data series with better quality for
decision-making systems. The method focuses on sensor fusion in order to obtain
inferences about the environment in a complementary way, trying to infer
information from different sensor sources (different properties) in a dynamically
changing environment. The employment of sensor fusion methods is to provide
agricultural and environmental monitoring a better understanding of the place
under observation. Sensor Fusion methods can abstract information from raw data
collected by the sensor nodes in order to show the environment real condition.
The Adaptive Sensor Fusion method proposed can be set to behave adaptively to
input stimuli and adjust to changes in real time seeking better results for decisionmaking systems. By the application of the method proposed, the error rate is
expected to be reduced by making applications based on these data more reliable.
Keywords: Wireless Sensor Network,
Agriculture, Water Quality.
Adaptive Technology, Precision
INTRODUCTION
Wireless Sensor Network (WSN) is a sub-class of ad hoc networks and it has
been used widely in two application categories: monitoring (closed and open
environments, health and wellness, process automation, seismic and structural)
and location (objects, animals, people and vehicles) (GAJBHIYE; MAHAJAN,
2008).
In the case of agricultural environment applications, its use provides better
coverage and control of the monitored area, because sensor nodes can be spread
out in the field, covering large areas. The requirement of large areas cover
requires large scale networks composed of many sensor nodes that monitor the
environment continuously or at predetermined time intervals, consequently
resulting in a large volume of data collected.
In environmental monitoring applications, for example, evaluating the water
quality of rivers can be made by WSN, which allows a faster response than the
conventional way, which requires manual collection of water samples and
laboratory analysis. Hence, one of the advantages in using WSN is the possibility
of real-time monitoring, allowing to take action more promptly to problems
encountered.
MOTIVATION
Sensor Fusion (SF) mechanisms can be employed to evaluate a large volume
of data collected by WSN. According to Llinas and Hall (1997), sensor fusion is
the combination of data from various sensors (with related information provided
by associated databases) to obtain better accuracy and more specific information
than the inferences achieved by using data from a single sensor. In agricultural
monitoring, environmental changes occur dynamically during the data collection
process, the Sensor Fusion mechanism being appropriate to adapt to these
changes; consequently, fine results can be achieved.
Thereby, this work proposes the design of a SF model with characteristics of
an Adaptive Device (AD), designed in accordance with the concepts of Adaptive
Technology (AT). Neto (2001) defines an AD as a formalism that has the ability
to dynamically change its topology and its behavior autonomously, selfmodifying to detect situations that require changes in response to input stimuli.
This research focus is the design of an Adaptive Sensor Fusion Model (ASFM)
responsible for evaluating and analyzing data collected by WSN in agricultural
and environmental monitoring applications, providing reliable information to
decision-making systems, trying to reduce error rates in these systems. The
proposal to provide adaptively to a sensor fusion model aims to contribute
adjusting its operation rules in accordance with the changes that may occur in the
environment or at the sensor network, such as failures or loss of some sensors.
Thus, the lack of data from sensors that may fail temporarily should have the
minimum impact possible on the monitoring purpose.
Another intention of this model is the use of sensor fusion to infer information
that could not be collected from the raw data, using a fusion of different variables.
This new information will aid decision making, be it in the activation of some
action or issue alerts, for example.
METHODS
The ASFM seeks to implement sensor fusion at different levels of data
processing. As shown in Figure 1 (an example of WSN configuration), nodes are
organized in clusters, the cluster head holding a first level of SF. At point A, SF
mechanisms would be employed in order to improve data quality, eliminating
false readings, estimating values in temporary failure of any sensor, and aggregate
data to reduce traffic to the root cluster head. Examples of mechanisms to be
considered: Maximum Likelihood Estimator, Least Squares, Kalman Filter. At
point B, the root cluster head accounts for implementing the data merger to infer
and abstract information to improve the performance of the identification process
of the present situation, and to support decision making. Examples of mechanisms
to be used have been: Bayesian inference, Dempster-Shafer Theory, Fuzzy Logic
and Neural Networks.
In Figure 1, each cluster consists of five sensor nodes and one cluster head,
responsible for sensor fusion to improve the quality of collected data. The root
accounts for inference, and can communicate with other mobile devices or
servers.
AT can work both point A and point B, making the input selection of SF
mechanism and dynamically adjust actions in accordance with the available
sensors and other environmental changes, as shown in Figure 2.
Figure 1 - Adaptive Sensor Fusion Model: initial proposal. Based on
(BITENCORT, 2007).
The model is based on fusion models oriented action, such as the Waterfall
Model. The complexity of the subtasks is increased, from the sensor data to the
feature extraction and decision making (RAOL, 2009).
Figure 2 - Proposed model main components.
RESULTS
The proposed model is still at an early stage of defining which SF mechanisms
AT to use. The next study involves the detailing of the model, as well as defining
mechanisms for TA to be included in the mechanisms of SF seeking better data
quality. The intention is to conduct an experiment in environmental monitoring,
evaluating water quality in rivers, offering a real-time monitoring, exploring the
application of SF to generate new levels of water quality, based entirely on data
collected by sensors in a WSN.
Some applications in the literature show the use of SF mechanisms in
agricultural equipment (instrumentation), and it is not easy to find SF applications
with data collected via WSN. Sun et al. (2007) developed a SF system that, using
data collected by sensors, allows estimating soil water retention, mechanical
strength and electrical conductivity in real-time monitoring. The authors state that
this system, combined with location data (GPS), can provide further valuable
information on soil properties for the AP (SUN; ZENG, ZHU, 2007). Some
examples of SF mechanisms applications, together with WSN are presented in
precision irrigation, as the experiment described by Xiong and Wang (2009). In
their experiment, the authors present the application of certain technologies in
WSN for precision irrigation based on the acoustic signals monitoring of water
stress in the crop using SF to improve data accuracy and to ensure reliable
decision making.
ACKNOWLEDGMENTS
The authors thank the Foundation for Research Support of the State of Mato
Grosso (FAPEMAT) for supporting this work (project nos. 278718/2010 and no.
313438/2011), and the Coordination of Improvement of Higher Education
Personnel (CAPES) via the inter-institutional Ph.D. program with the Engineering
School at University of São Paulo and the Federal University of Mato Grosso
(UFMT). Thanks also to Professor Dr. D.S. Barbosa (UFMT) for the
collaboration to work related to the environmental and agricultural area.
CONCLUSION
The need of computational methods, such as Sensor Fusion for the data
processing generated by sensor nodes from a WSN arises, among other factors,
from the demand for information. Based on reliable information, decision-making
by managers may follow more elaborate technical criteria. SF mechanisms can be
applied to data collected by the WSN in order to reduce data inconsistency. The
environmental damage generated by agricultural production, especially water
contamination, needs monitoring tools that allow a continuous, faster and more
effective automated decision making system. WSN has thus great potential when
used in solutions for environmental quality evaluation.
The Adaptive Sensor Fusion Model proposed for water quality monitoring is
intend to infer the possibility of new quality indices that use fewer sensors, low
cost (less variable) providing a similar level of reliability of existing IQA.
REFERENCE
BITENCORT, B.; CABREIRA, U.; DANTAS, M.A.R.; CARLOS, M.; PINTO,
A. 2007. Probabilistic Real-time Data Fusion in Wireless Sensor Networks
with ZigBee. In: 7th IFAC International Conference in Fieldbuses and
Networks in industrial Embedded Systems. (Toulouse: 2007). Vol. 1, pp.
267-271.
GAJBHIYE, P.; MAHAJAN, A. 2008. A survey of architecture and node
deployment in Wireless Sensor Network. Applications of Digital Information
and Web Technologies, 2008. ICADIWT 2008. First International
Conference on the, 2008. 4-6 Aug. 2008. pp.426-430.
HALL, D. L.; LLINAS, J. An introduction to multisensor data fusion.
Proceedings of the IEEE, v. 85, n. 1, pp. 6-23, 1997. ISSN 0018-9219.
NETO, J.J. Adaptive Rule-Driven Devices – General Formulation and Case
Study. 6th International Conference, 2001. Pretoria, South Africa: SpringerVerlag. pp.234-250.
RAOL, J.R. 2009. Ed. Multi-Sensor Data Fusion with MATLAB. Bangalore:
CRC PRESS.
SUN, Y.; ZENG, Q.; ZHU, Z. 2007. Measuring Soil Physical Properties by
Sensor Fusion Technique. Industrial Electronics and Applications, 2007.
ICIEA 2007. 2nd IEEE Conference on, 2007. 23-25 May 2007. pp.142-146.
XIONG, S.M.; WANG, L.M.; QU, X.Q.; ZHAN, Y.Z. 2009. Application
Research of WSN in Precise Agriculture Irrigation. Environmental Science
and Information Application Technology, 2009. ESIAT 2009. International
Conference on, 2009. 4-5 July 2009. pp.297-300.